Claude Sonnet vs Opus: Which One to Use

Claude Sonnet vs Opus comes down to capability, speed and price. Claude Opus 5.5 is the more capable model and costs twice as much as Claude Sonnet 5.5: $4 per million input tokens and $20 per million output tokens, against $2 and $10 for Sonnet. Sonnet is faster. Anthropic suggests starting most work on Opus 5.5, and Claude Code already uses Opus 5.5 by default on every paid plan. Switch to Sonnet when speed or cost matters more than the last bit of capability.

Everything below comes from Anthropic's model docs and the Claude Code model settings, checked November 7, 2026. Anthropic ships new models often, so check those pages if the version numbers have moved on.

Sonnet vs Opus at a glance

What's actually different

Capability. Opus is the larger model. Anthropic positions it for multihour autonomous coding, large refactors, complex systems engineering, vision-heavy workflows and computer use. Sonnet is pitched at code generation, data analysis, content creation, visual understanding and agentic tool use. Both handle text and images, tool use and many languages.

Speed. Anthropic rates Sonnet as fast and Opus as moderate. Real speed depends on how long your prompt is, how much the model writes and the effort level you choose.

Price. Opus costs exactly twice as much at every step: input, output, cache writes and cache reads. Both get 50% off through the Batch API.

Thinking. Both use adaptive thinking, which means the model decides how much to think on each step. On Opus 5.5 it's always on and can't be turned off. On Sonnet 5.5 you can turn off thinking up front, so it starts working straight away, which helps when you want quick replies.

What's the same. Both have a 1M token context window, 128K tokens of output, the same June 2026 knowledge cutoff and the same five effort levels: low, medium, high, xhigh and max.

What the price gap looks like

Here's one large job worked through at list prices, before caching:

On a Pro, Max or Ultra plan in Claude Code you don't pay per token, so the price gap shows up as how far your plan's usage goes rather than as a bill. If you pay through the API or a Console account, it's a straight 2x. Our Claude Code pricing guide covers both ways of paying.

When to use Sonnet

Pick Sonnet 5.5 when the task is clear and you want it done fast:

• Writing a function, a test or a small feature you've described well
• Editing copy, emails or docs
• Pulling data out of files and turning it into tables
• Agents in production that run many times a day, where cost adds up
• Subagents that search files or run tests for a bigger session

If you're building something for a client, Sonnet is often the right default for the live system. A support agent or a lead-qualifying agent that runs thousands of times a month costs half as much on Sonnet. Test both models on 20 or so real examples from the client before you choose.

When to use Opus

Pick Opus 5.5 when getting it wrong costs more than the extra tokens:

• Work that touches many files or a whole codebase
• Debugging where nobody knows the cause yet
• Planning an architecture or a migration
• Long sessions where Claude works on its own for hours
• Tasks that need it to read screenshots, diagrams or a screen

Anthropic's own advice for its developer platform is to start with Opus 5.5 for most workloads, then move to a cheaper model or a lower effort level once you know what the task needs.

Where Fable and Haiku fit

Sonnet and Opus sit in the middle of the lineup. Claude Fable 5.1 sits above them for the hardest reasoning and long-running agent work, at $10 and $50 per million tokens. Claude Haiku 5.5 sits below them for high-volume tasks like sorting, extraction and routing, from $0.10 and $0.50.

How to switch between Sonnet and Opus in Claude Code

Claude Code starts on Opus 5.5 on Pro, Max, Ultra, Team and Enterprise plans and on Anthropic API accounts. To change it during a session, run:

/model sonnet

Switch back with /model opus, or run /model on its own to open the picker. To start a session on a model, launch it with claude --model sonnet. The picker saves your choice for new sessions when you press Enter, or for this session only when you press s.

There's also a hybrid setting:

/model opusplan

It uses Opus while you're in plan mode and switches to Sonnet when Claude starts writing code. That gets you Opus for the thinking and Sonnet's speed and cost for the typing.

Two more things to know:

• Subagents follow your session's model unless their definition sets one. Put model: sonnet in a custom subagent to keep its routine work on the cheaper model.
• Effort is a separate setting. In Claude Code, Opus 5.5 and Sonnet 5.5 both start at medium. Change it with /effort.

Model or effort: which one to change

When Claude gets something wrong, you have two dials. Anthropic's Claude Code team gives a simple test in its post on choosing a model and effort level:

Effort controls how much work Claude does on a turn, including how many files it reads and how much it checks. The model controls what it's capable of. So a higher effort on Sonnet can fix a lazy answer, but it won't make Sonnet as capable as Opus.

Using Sonnet and Opus through the API

Use claude-sonnet-5-5 or claude-opus-5-5 as the model ID. Both are on the Claude API, Amazon Bedrock, Google Cloud, Microsoft Foundry and Claude Platform on AWS. Two API details catch people out:

• The default effort differs. Sonnet 5.5 defaults to high and Opus 5.5 to medium, so set it yourself if you compare them.
• Sonnet 5.5 returns an error if you set temperature, top_p or top_k. Remove those settings when you switch an older app over.

Which should you pick?

Start on Opus if you're unsure, because it's the default and it makes fewer mistakes on hard work. Drop to Sonnet for routine tasks, quick edits and anything that runs at volume. Try opusplan if you want both in one session.

New to Claude Code? Start with how to install it, then see how it stacks up in Claude Code vs Codex. To turn these skills into paid client work, the Claude Code for agencies course walks through it, and members compare setups in the Claude Code community.

FAQ

Is Claude Opus better than Sonnet?

Opus 5.5 is more capable than Sonnet 5.5, especially on long, complex coding and multistep work. Sonnet is faster and costs half as much, and for clear, routine tasks you often won't see a difference in the result.

How much more expensive is Opus than Sonnet?

Twice as much. Opus 5.5 costs $4 per million input tokens and $20 per million output tokens. Sonnet 5.5 costs $2 and $10. Cache prices are also double on Opus.

Which model does Claude Code use by default?

Opus 5.5 on Pro, Max, Ultra, Team and Enterprise plans and on Anthropic API accounts. Run /model sonnet to switch to Sonnet.

What is opusplan in Claude Code?

A model setting that uses Opus while you're in plan mode and Sonnet when Claude carries out the plan. Turn it on with /model opusplan.

Do Sonnet and Opus have the same context window?

Yes. Sonnet 5.5 and Opus 5.5 both have a 1M token context window and can write up to 128K tokens in one response.

Should I use Sonnet or Opus for an AI agent I build for a client?

Test both on real examples first. If Sonnet gets the answers right, use it, because the agent will run many times and Sonnet costs half as much. Keep Opus for agents that make complex decisions or work across many steps.